finetune.py

#!/usr/bin/python3
from transformers import T5Tokenizer, T5ForConditionalGeneration, Trainer, TrainingArguments
from datasets import load_dataset

# Load model and tokenizer
model_name = "t5-small"  # or another transformer-based model
tokenizer = T5Tokenizer.from_pretrained(model_name)
model = T5ForConditionalGeneration.from_pretrained(model_name)

# Load dataset
dataset = load_dataset("json", data_files={"train": "train.json"})  
evalset = load_dataset("json", data_files={"eval":  "eval.json"})

def preprocess_function(examples):
    inputs = ["Generate a question for: " + (ans if isinstance(ans, str) else "Unknown") for ans in examples["answer"]]
    model_inputs = tokenizer(inputs, max_length=512, truncation=True, padding="max_length")  # <-- Added padding

    labels = [q if isinstance(q, str) else "" for q in examples["question"]]
    labels = tokenizer(labels, max_length=128, truncation=True, padding="max_length")  # <-- Added padding

    model_inputs["labels"] = labels["input_ids"]
    return model_inputs

tokenized_datasets = dataset.map(preprocess_function, batched=True)
tokenized_evalsets = evalset.map(preprocess_function, batched=True)

# Define training arguments
training_args = TrainingArguments(
    output_dir="./results",
    evaluation_strategy="epoch",
    save_strategy="epoch",
    per_device_train_batch_size=8,
    per_device_eval_batch_size=8,
    num_train_epochs=3,
    weight_decay=0.01,
    logging_dir="./logs",
)

trainer = Trainer(
    model=model,
    args=training_args,
    train_dataset=tokenized_datasets["train"],
    eval_dataset=tokenized_evalsets["eval"]
)

# Train model
trainer.train()

# Save trained model
output_dir = "./aq_model"  # Change the folder name if needed
trainer.save_model(output_dir)
tokenizer.save_pretrained(output_dir)

print(f"Model saved to {output_dir}")